Data Visualization with R

• About this course

• Acknowledgements

• Introduction to Graphics
• Base ggplot2
• Creating a Scatter Plot
• Creating a Line Plot
• Creating a Bar Plot
• Creating a Histogram
• Creating a Box Plot
• Creating a Surface Plot
• Advanced ggplot2
• Mapped Aes Customization
• Axes Customization
• Legend Customization
• Facet Customization
• Plot Overall Appearance
• More on stats
• Other topics
• ggplot2 community
• Presenting Plots

Mapped Aesthetics Customization

require(dplyr)
require(ggplot2)
require(qdata)
data(bands)

You can change the appearance of mapped aesthetics by using scales.
Scales control the mapping from data to aestethics. They take your data and turn into something that you can see like colour, shape, position or size. Scales provide also the tools that allow you to read the plot: legends and axes.

In this chapter we will discuss how customize the appearance of mapped aesthetics by exploring scales functionalities. Legends and axes will not discussed here but in Axes Customization and Legend Customization chapters, respectively.

But firstly, let us introduce some concepts about scales structure.

Scales Structure

A scale is required for every aestethic used on a plot. For example, when you write:

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity)) +
  geom_point(mapping = aes(colour=band_type))

What actually happens is:

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity)) +
  geom_point(mapping = aes(colour=band_type)) +
  scale_x_continuous() +
  scale_y_continuous() +
  scale_color_discrete()

The previous two specifications are equal.

Default scales are named according to the aesthetic and the variable type. It would be tedious to manually add a scale every time a new aestethic is added, so ggplot2 does it for you.

Each scale has default values for its argument, this means that if you want to modify one or more features of this scale you have to overwrite it/s default value. To overwrite its default value/s you have to manually add that scale.

For example, let us suppose that you want to reduce the luminance of the points colours in the plot. Colour aesthetic mapping is controlled by scale_color_discrete() and the argument of scale_color_discrete() that control the luminance is l. l can assume values from 0 to 100 and its default value is 65. We want to decrease the luminance and set it to 40 in this way:

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity)) +
  geom_point(mapping = aes(colour=band_type)) +
  scale_color_discrete(l = 45)

The use of + to “add” a scale is a little misleading. Remember that when you + a scale you are not adding it but ovverwriting it.

Scales naming scheme and scales types

The name for scaling scheme is made up by three pieces separated by _:

1. scale
2. the name of the aestethic (e.g., colour, shape, …)
3. the name of the scale (e.g., continuous, discrete, brewer, …)

There are more scale types. They can be grouped into:

• Continuous position scales, used to map data into x and y position
• Mapped aestethics scales, like colour, linetype, shape and size.

In the following paragraphs we will answer to the most frequently asked questions about mapped aesthetics scales customization for: colours, linetype, shape and size. The use and the features of continuous position scales will be discussed in Axes Customization chapter.

Customizing colours

Do you know something about colour theory? If the answer is no, you should read this paragraph.

At the physical level, colour is produced by a mixture of wavelenghts of lights. To characterise a colour completely we need to know the complete mixture of wavelenghts. Human eye has only three different colour receptors, so the perception of any colour can be summarised with just three numbers. You may be familiar with the RGB encoding of colour space, which defines a colour by the intensities of red, green and blue light needed to produce it. One problem with this space is that it is not percentually uniform and this make it difficult to create a mapping from a continuous variable to a set of colours. There have been many attemps to come up with colours spaces that are more percentually uniform.

We use a modern attemp called hcl colour space, which has three components of h (hue), c (chroma) and l (luminance):

• hue is a number between 0 and 360 (an angle) which gives the “colour” of the colour: like blue, red, orange, etc.
• chroma is the purity of a coloyr. A chroma of 0 is grey, and the maximum value of chroma varies with luminance
• luminance is the lightness of the colour. A luminance of 0 produces black, and a lumimance of 1 produces white

Hues are not perceived as being ordered, but chroma and luminance are ordered.

Change the colors for a discrete mapped variable

Suppose we are interested in the relationship between humidity and viscosity by press_type in bands dataset:

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point()

You want to use different colour for press_type levels so you want to change the default palette. A palette is a range of colours; so changing a palette means to modify the colour (or fill) scale: it involves a change in the mapping from numeric or categorical values to aesthetic attributes.

There are two types of scales that use colors: fill scales and color scales.
You can change the default colours used in the graph by choosing a function from the following table:

Fill scale	Colour scale	Description
scale_fill_discrete()	scale_colour_discrete()	Colors evenly spaced around the color wheel (same as hue)
scale_fill_hue()	scale_colour_hue()	Colors evenly spaced around the color wheel (same as discrete)
scale_fill_grey()	scale_colour_grey()	Greyscale palette
scale_fill_brewer()	scale_colour_grey()	ColorBrewer palettes
scale_fill_manual()	scale_colour_manual()	Manually specified colors

By default, the colors for discrete scales are evenly spaced around a HSL color circle.

For example, if there are two colors, then they will be selected from opposite points on the circle; if there are three colors, they will be 120° apart on the color circle; and so on. The default colors used for different numbers of levels are shown here:

The default color selection uses scale_fill_hue() and scale_colour_hue():

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_color_hue()

The previous two graph are equal and in the second graph definition scale_color_hue() is redundant.

With scale_colour_hue(), the colors are taken from around the color wheel in the HCL (hue-chroma-lightness) color space. In “Scales structure” we learn that the default lightness value is 65 on a scale from 0–100. This is good for filled areas, but it’s a bit light for points and lines. To make the colors darker for points and lines, set the value of l (luminance/ lightness):

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_colour_hue(l=45)

You can also use other color scales, such as ones taken from the RColorBrewer package:

The ColorBrewer palettes can be selected by name:

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_colour_brewer(palette="Set1")

You can also use a palette of greys. This is useful for print when the output is in black and white.

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_colour_grey()

The default is to start at 0.2 and end at 0.8, on a scale from 0 (black) to 1 (white), but you can change the range:

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_colour_grey(start=0.7, end=0)

Using a Manually Defined Palette for a Discrete Variable

Colour can be also manually specified by using scale_colour_manual() function. The colors can be named, or they can be specified with RGB values:

# named colours
ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_colour_manual(values = c("magenta", "dark turquoise", "dodger blue", "lime green"))

# RGB colour code
ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_colour_manual(values = c("#FF00FF", "#00CED1", "#1E90FF", "#32CD32"))

The order of the items in the values vector matches the order of the factor levels for the discrete scale.

levels(bands$press_type)

## [1] "ALBERT70"  "MOTTER70"  "MOTTER94"  "WOODHOE70"

In the preceding example, the order of press_type is “ALBERT70”, then “MOTTER70”, then “MOTTER94” and “WOODHOE70”, so the first in values argument goes with “ALBERT70”, the second goes with “MOTTER70”, the third with “MOTTER94” and the fourth with “WOODHOE70”. If the variable is a character vector, not a factor, it will automatically be converted to a factor, and by default the levels will appear in alphabetical order.

It’s possible to specify the colors in a different order by using a named vector:

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = press_type)) +
  geom_point() +
  scale_colour_manual(values = c(ALBERT70 = "lime green", MOTTER70 = "dark turquoise", MOTTER94 = "magenta", WOODHOE70 = "dodger blue"))

There is a large set of named colors in R, which you can see by running colors(). Some basic color names are useful: "white", "black", "grey80", "red", "blue", "darkred", and so on. There are many other named colors, but their names are generally not very informative, for example "thistle3" and "seashell", so it is often easier to use numeric RGB values for specifying colors.
RGB colors are specified as six-digit hexadecimal (base-16) numbers of the form "#RRGGBB". In hexadecimal, the digits go from 0 to 9, and then continue with A (10 in base 10) to F (15 in base 10). Each color is represented by two digits and can range from 00 to FF (255 in base 10). So, for example, the color "#FF0099" has a value of 255 for red, 0 for green, and 153 for blue, resulting in a shade of magenta. The hexadecimal numbers for each color channel often repeat the same digit because it makes them a little easier to read, and because the precise value of the second digit has a relatively insignificant effect on appearance.

Here are some rules of thumb for specifying and adjusting RGB colors:

• higher numbers are brighter and lower numbers are darker
• to get a shade of grey, set all the channels to the same value

Using a Manually Defined Palette for a Continuous Variable

You want to use different colors for a continuous variable.

Mapping continuous values to a color scale requires a continuously changing palette of colors.
The following table lists the continuous color and fill scales:

Fill scale	Colour scale	Description
scale_fill_gradient()	scale_colour_gradient()	Two-color gradient
scale_fill_gradient2()	scale_colour_gradient2()	Gradient with a middle color and two colors that diverge from it
scale_fill_gradientn()	scale_colour_gradientn()	Gradient with n colors, equally spaced

Supposing we are interested in the relationship between humidity, viscosity and solvent_pct in bands dataset:

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = solvent_pct)) +
  geom_point()

Let us see how the previous fuctions work. We will specify the colours for a continuous variable using various gradient scales. The colors can be named, or they can be specified with RGB values:

# scale_colour_gradient()
ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = solvent_pct)) +
  geom_point() + 
  scale_colour_gradient(low="black", high="white")

# scale_colour_gradient2()
ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = solvent_pct)) +
  geom_point() + 
  scale_colour_gradient2(low="red", mid="white", high="blue")

# scale_colour_gradientn()
ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = solvent_pct)) +
  geom_point() + 
  scale_colour_gradientn(colours = c("darkred", "orange", "yellow", "white"))

You can also use palettes produced by another package:

ggplot(data = bands, mapping = aes(x = humidity, y = viscosity, colour = solvent_pct)) +
  geom_point() + 
  scale_colour_gradientn(colours = terrain.colors(n=8))

In this case we use terrain.colors palette with 8 colours. terrain.colors is a palette of grDevices package.

Customizing linetype

R provides the following line types:

linetype aesthetic has to be mapped to discrete variables.

Changing line types

Let us consider the average growth of chicks for the different diets by using ChickWeights dataset, as seen in Creating a Line Plot:

# data
ChickWeightMean <- ChickWeight %>% 
  group_by(Time, Diet) %>% 
  summarize(weight=mean(weight)) %>%
  mutate(Diet = factor(Diet, levels=1:4, labels=c("A","B","C","D")))

We want to se the difference in the average growth of chicks for the different diets, by mapping Diet variable to linetype aestethics:

ggplot(data=ChickWeightMean, mapping=aes(x=Time, y=weight)) +
  geom_line(mapping=aes(linetype=Diet)) 

To change linetypes we have to set values argument of scale_linetype_manual() function:

ggplot(data=ChickWeightMean, mapping=aes(x=Time, y=weight)) +
  geom_line(mapping=aes(linetype=Diet)) +  
  scale_linetype_manual(values=c(3,4,5,6))

Line type values can be passed as numbers or strings.

Customizing Shape

R provides the following shapes symbols:

shape aesthetic has to be mapped to discrete variables.

Changing shape types

If you want to customize shape scale you have to set values argument of scale_shape_manual() function:

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity, shape=press_type)) +
  geom_point() +
  scale_shape_manual(values = c(12, 13, 8, 3))

Customizing Size

The size aesthetic is most commonly used for points and text, and humans perceive the area of points (not their radius), so this provides for optimal perception. size can be mapped only to continuous variables.

Change the size of plotting symbols

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity, size=ink_pct)) +
  geom_point()

When a variable is mapped to size, the results can be perceptually misleading. If it is important for the sizes to proportionally represent the quantities, you can change the range of sizes. By default the sizes of points go from 1 to 6 mm. You could reduce the range to, say, 2 to 4 mm, with scale_size_continuous(range=c(2, 4)).

If you want to customize the minimum and the maximum size of the plotting symbol you have :

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity, size=ink_pct)) +
  geom_point() +
  scale_size(range=c(2, 4))

So you have to set the to range argument the minimum and the maximum size of the plotting symbol.

However, the point size numbers don’t map linearly to diameter or area, so this still won’t give a very accurate representation of the values.

If you want that a value of zero is always mapped to size 0, use scale_size_area() instead of scale_size():

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity, size=ink_pct)) +
  geom_point() +
  scale_size_area()

However with scale_size_area() you can not control the size of plotting symbols by using range argument. You can set only the size of largest points by setting max_size argument:

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity, size=ink_pct)) +
  geom_point() +
  scale_size_area(max_size = 3)

You can also use radiuos to control the size of plotting symbols by setting scale_radius() function:

ggplot(data=bands, mapping=aes(x=humidity, y=viscosity, size=ink_pct)) +
  geom_point() +
  scale_radius()

This choice is always not recommended.